Wearable-Assisted Localization and Inspection Guidance System Using Egocentric Stereo Cameras

Abstract

As industrial inspection and maintenance tasks in outdoor working environments are increasingly complex and flexible, wearable assistive system (WAS) has shown promising prospects, especially when the tasks are infrequent and unfamiliar to workers. Such tasks require more accurate and global ego-localization than using GPS. In this paper, a wearable-assisted navigation system is proposed for task-specific route guidance. This paper proposes to learn a global metric-topological abstract map of outdoor workspace using the combination of stereo cameras and inertial measurement units. An extended stereo visual simultaneous localization and mapping approach is proposed with optimization that deals with cross-session loop-closure detection and pose rectification for bounding estimation error as well as aligning the positions to global world frame. Thus, the system allows building and using globally consistent maps across sessions. A prototype of WAS is developed in which the augmentations of guidance are delivered to users via an optical see-through display. The system is evaluated by experiments designed as on-site inspection scenarios and the results validate the practicability and effectiveness of the system.